GiveDirectly - April 2012 Review

We have published a more recent review of this organization. See our most recent report on GiveDirectly.

Published: April 2012

Our review process

Our evaluation of GiveDirectly consisted of reviewing publicly available information, speaking by phone and meeting with GiveDirectly's founders, and reviewing information that GiveDirectly sent to us in response to our questions.

Older content on GiveDirectly:

• July 2011 blog post on GiveDirectly

What do they do?

GiveDirectly transfers cash to poor households in Kenya via the M-Pesa mobile phone-based payment service.

GiveDirectly is a small, new organization. It started accepting donations from the public in 2011, and has had one full-time staff member since January 2011.1 As of April 2012, GiveDirectly had spent $166,000, of which $130,000 went to grants to 900 households (see below; note that numbers in the table below do not match aggregate figures because the numbers are from slightly different time periods). GiveDirectly's total expenses to date include startup costs that are not likely to recur in the future.

The table below shows all of the grants that GiveDirectly has made (or committed to make) in its existence, the number and type of recipient, the amount the recipient will receive from GiveDirectly, the percent of grants of this type paid so far as of March 31, 2012, the total granted so far and still to be granted, and the dates during which the transfer will take place. GiveDirectly plans to make future transfers as two yearly installments of $500;2 for RCT participants, GiveDirectly is "currently experimenting with different payment schedules, including monthly payments and larger one-time transfers."3

Targeting of households

GiveDirectly's targeting system:5

1. First select areas in which to work based on poverty statistics6
2. Then, using unannounced field visits, select households within these areas who live in "homes made from organic materials (lacking cement, stone or iron floors and walls and without an iron roof)."7

Transferring of funds

GiveDirectly transfers funds to recipients' mobile phones through a national, independent money transfer service, M-Pesa. GiveDirectly gives SIM cards to recipients who do not have mobile phones.8 Funds can then be sent to the recipient's SIM card and the recipient can redeem the balance for cash at a local agent.9

Does it work?

We believe that giving cash directly to the poor is a promising way to help people (see our report on cash transfers). Intuitively speaking, it seems that cash grants (if appropriately directed) allow recipients to use money as they see fit, meeting the needs individuals feel rather than needs disconnected organizations may perceive for them. In order to determine whether a cash transfer program is effective, we ask:

1. Are the intended recipients poor?
2. Do funds reach the intended recipients (rather than being misappropriated by more powerful interests)?
3. Are grant sizes appropriately targeted to the needs of recipients?
4. Do grants cause unintended negative consequences such as jealousy or market distortions?

We also ask: do the transfers raise recipients' standards of living above and beyond the value of the transfer? In order to compare GiveDirectly's program to other types of programs, such as delivering proven medical treatments, we seek to fully understand the benefits of cash transfers.

GiveDirectly is a small, new organization. Below, we describe both (a) the processes GiveDirectly has said it will apply to its future activities to answer the above questions, and (b) early results from GiveDirectly's work to date. Importantly, the early results we have seen are from transfers made to participants in a randomized controlled trial (RCT) that GiveDirectly is running, because these were among the first households to receive transfers from GiveDirectly. These recipients differ from planned future recipients in that (a) some of them are receiving significantly smaller grants ($300 during the study period instead of $1000) and no recipients have yet received over $300,10 and (b) they live in areas where some households that qualify for transfers under GiveDirectly's targeting criteria are not currently receiving transfers (because they were selected to be in the control group). As a result, findings about the community effects of the transfers may not be generalizable to future transfers. In addition, there is some additional data collection among households participating in the RCT.

Are intended recipients poor?

GiveDirectly's targeting methodology is to select recipient households based on three observable characteristics: the materials of their roof, walls, and floor. All must be made of organic materials for the household to qualify.11 GiveDirectly believes that households that have iron roofs or homes made of cement are generally richer than those with homes made without using any of these materials (also referred to by GiveDirectly as "mud and thatch" homes).12 We have two main questions about whether this methodology results in targeting of very poor households:

• Are most households that live in houses made of organic materials very poor? As part of the research conducted for the RCT, GiveDirectly has collected data on poverty levels of RCT participants, who were selected for participation based on the housing material criterion. GiveDirectly has shared the full survey form used to interview participants.13 GiveDirectly reports that recipients are very poor:14
  
  Well over half of adults skip meals, less than half of household members eat until they are content, people commonly go to sleep hungry and a paltry 18% report having enough food for tomorrow in their household. Those living in eligible homes are even worse-off than the average household, consuming less and holding fewer assets. Overall, mean and median daily per capita consumption among eligible households are $0.65 and $0.55 at nominal rates, and 74% are below the Kenyan poverty line, indicating a very poor population.

  GiveDirectly has offered to share the full results of this survey with us; we have not yet reviewed these results.

  GiveDirectly told us that going forward it will measure a smaller number of indicators of poverty status during its enrollment process. As of February 2012, it had not yet selected which indicators it will measure and plans to do this based which indicators "are good statistical predictors of broader measures of consumption poverty" in the RCT baseline study.15

  It is not clear from this survey that households excluded from the program are significantly less poor. GiveDirectly provides a chart that shows that a clear difference between households in mud and thatch homes and those in cement homes, but fairly small differences between those living in mud and thatch homes and those living in mud and iron sheet homes.16 GiveDirectly told us that households were randomly selected for participation in the survey; we have not seen further technical details from the study.17

• Is the targeting criteria applied consistently (i.e. do staff select only households living in houses made of organic materials)? To monitor whether selection of recipients is properly carried out by field staff, GiveDirectly's process is to conduct two rounds of verification:18
  • First, a second field officer visits each selected household to check that it meets the eligibility requirements. Our understanding is that GiveDirectly has carried out second visits to households that have received funds. Note that we have not seen documentation from these visits because GiveDirectly told us that these documents contain identifying information about individual recipients and sharing them could raise privacy concerns.19
  • Second, a foreign employee or board member audits at least 10% of selected households. Audits include checking the eligibility of the household and asking whether the recipient had been asked to pay a bribe.20 GiveDirectly sent us a report on the 140 audits completed to date. GiveDirectly reported one case of "borderline eligibility" in the 140 audits and no reports of bribes.21 The household found to be of "borderline eligibility" did not receive funding from GiveDirectly.22 GiveDirectly has provided a list of villages that were visited during audits, who conducted the audit, the date of each visit, and the number of households visited in each village.23 GiveDirectly told us, "Villages were selected [for audit] on a convenience basis (typically a supervisor whose workflow slackened would take a day to visit the most recently enrolled village). Within these villages all the enrolled households were audited."24 GiveDirectly told us that it audited a high percentage of the first sets of households it transfered funds to and that as it scales up it plans to audit a lower percentage of households than it has in the past.25

Do funds reach intended recipients?

GiveDirectly plans to follow up with recipients to determine whether they received their full transfers. GiveDirectly describes this process:26

Early results: In November 2011, GiveDirectly shared with us its first round of follow-up survey data from 174 households (these are households that are receiving transfers as part of GiveDirectly's RCT). GiveDirectly shared the list of questions that it used and the full data set.27 GiveDirectly notes that only "successful surveys" were included; participants who could not be reached by phone were not included in the results.28 As of February 2012, GiveDirectly was considering visiting a sample of households that had not been reached by phone in-person. GiveDirectly provided data on the response rate through March 2012 (a longer time frame than is covered by the results we have seen). As of this date, GiveDirectly had succeeded in following up with 87% of all recipients.29

At the time of the follow-up survey, 89% of participants reported receiving transfers, 3% reported that the money had been transferred to them but they had not collected it yet, and 7% reported not having received the transfer. Of those who did not receive their transfers the issue in all but one case was that the recipient had not yet been registered with M-Pesa. GiveDirectly was later able to confirm that all of these households had registered,30 and told us that it confirmed receipt of transfers with households via phone.31 No recipients reported being asked for a bribe to be enrolled in the program. Two reported being asked for a bribe from the M-Pesa agent when collecting the transfer (one later said that she had not paid a bribe and had misunderstood the question in the first follow up).32 On average, recipients reported paying 3.32 Ksh (about $0.04) more than the expected official M-Pesa withdrawal fee. 14 households reported difficulties withdrawing their transfers. The most commonly cited difficulty was M-Pesa agents running out of cash. One household reported withdrawing 22,000 Ksh (about $250) and then losing the remaining amount (about $50) due to a network problem, and another household reported sending someone to pick up the transfer and losing 750 Ksh in the process (about $9).33

Are grant sizes appropriately targeted to the needs of recipients?

GiveDirectly expects to provide $1000 transfers (over two years) to recipients who may live on only a dollar or two per day. GiveDirectly's earlier recipients received significantly smaller grants.34 GiveDirectly is "currently experimenting with different payment schedules, including monthly payments and larger one-time transfers, as part of our ongoing [RCT] evaluation."35

GiveDirectly told us that it selected a $300 grant size for the RCT based on funding available at the start of the study. It later decided to provide $1000 grants to a portion of the RCT participants, because it raised additional funds and came to believe there was value in studying the differences in how households benefited from the $300 versus the $1000 grants. It may alter the size of future transfers based on the results of the RCT.36

GiveDirectly gives the following rationale for the $1000 transfers (note: we have not vetted the below figures):37

Do the transfers raise recipients' standards of living above and beyond the value of the transfer?

It is plausible that extremely poor individuals may be caught in a "poverty trap" such that an infusion of cash would not only raise their income by the size of the transfer but allow them to invest the funds and raise their incomes by more than the size of the transfer.

We have not seen evidence that this is occurring for GiveDirectly's recipients.

We discuss what independent studies have show about the track record of cash transfer programs at our in-depth evidence review.

GiveDirectly is conducting a randomized controlled trial with households each receiving $300 or $1000 and will measure "impacts of transfers on household consumption, hunger and nutrition, the entrepreneurial activities of households, health (both physical and emotional) and whether children are in school." GiveDirectly has pre-announced the design of the study, a step we believe increases the credibility of a study (see further discussion on our blog).38 This study will look both at households receiving $300 total and $1,000; the latter group is more likely to be comparable to future recipients who will also receive grants of $1,000 per household.39 The study is scheduled to conclude in early 2013.40

GiveDirectly notes that its decision to set the grant size at $1,000 per household was based, in part, on its belief that $1,000 is "how much an average-sized recipient household with five members would need to invest in typical income-generating activities...in order to raise their annual per-capita incomes to the level of their least well off but ineligible neighbors."41 We have not vetted this claim.

In early results from the 174 follow up surveys completed as of November 2011 (see above), households reported the following uses for their grants (respondents could provide multiple answers):42

• Home improvements: 43%
• Food: 40%
• School: 17%
• Clothes: 11%
• Livestock: 9%
• Medical care: 6%
• Investment in non farm business:43 6%
• Farm: 5%
• Buy land: 1%

Possible negative/offsetting impact

• Does distribution to some members of a community create jealousy among other members of the community?

  As discussed above, GiveDirectly conducts follow up interviews with recipients after transfers are sent. As part of the survey, GiveDirectly asks:44

  • Have there been any complaints or upset people in your community related to these transfers? Of the people who were complaining or upset, who were they upset with? Who has been complaining or upset about the transfers?
  • Has there been any shouting or angry arguments among people in your village about these transfers? If yes, describe.
  • Has there been any violence or crime in your village related to these transfers? If yes, describe.

  Of the 174 households surveyed to date, 14% reported complaints in the community, 2 households (1.2%) reported angry shouting or arguments, and 1 household (0.6%) reported violence of crime. Descriptions of the reported shouting/arguments and violence/crime are given in the full data set.45

  As far as we know, GiveDirectly does not ask non-recipient members of the community about the effect the program has had on them.

  As we note above, there appear to be fairly small differences in poverty level between recipients and many non-recipients, which could result in perceived unfairness.

• Do grants distort incentives for individuals?

  We have not seen information on the question of whether individuals who live in the areas served by GiveDirectly change their behavior in order to increase their chances of receiving transfers.

  It is possible that individuals could take measures to make themselves appear (or actually be) poorer in order to qualify for transfers, or spend more time at home (and perhaps less time working) to increase their chances of being at home when GiveDirectly staff visit. The one-off nature of transfers (recipients are not eligible for a second round of transfers),46 may help to mitigate these effects among past and current recipients.

• Are grants reliable?

  For all cash transfer programs, we are concerned that grants that are promised and not delivered or that arrive late could cause people to make plans and investments and be hurt by the absence or lateness of the transfer.

  After selecting recipients, GiveDirectly audits the eligibility of a sample and may decide not to make transfers to some households. GiveDirectly reports, "Of 140 such audits conducted to date, 1 (1%) detected a case of borderline eligibility"47 and the transfer was not made to that household.48

  GiveDirectly has interviewed some participants after transfers were sent to ask if they have received the transfers (GiveDirectly's policy is to interview all participants, but has only been able to follow up with about 55% of them to date49 ). In the data we have seen from 174 interviews, 13 participants (7.5%) reported that they had not received transfers yet. GiveDirectly reports that delays were primarily due to delays in recipients obtaining National ID cards (required for M-Pesa registration) and registering with M-Pesa. GiveDirectly reports from a later follow up with M-Pesa that all but 1 of the 13 had succeeded in registering and their transfers had been sent by this later date.50

• Do grants distort local markets?

  For all cash transfer programs, we are concerned about the ways in which large infusions of cash in an area could alter economic opportunities for both recipients and non-recipients. Such effects could be positive (for example by spurring investment and job creation or by increasing the availability of retail goods) or negative (for example by increasing demand and thus prices for goods in the area). It is not clear to us which effects are likely to dominate.

  GiveDirectly also seems to be concerned with this question. It has stated that, if it raises enough funds to do so (estimated in the range of several million dollars), it "intends to use a lottery to determine which among a set of similar villages receive transfers..., and then document the effects on the economic structure of 'treated' villages."51

What do you get for your dollar?

In our general review of cash transfer programs, we state that while we do not have information about the cost-effectiveness of cash transfers as compared to other interventions, our intuition is that our top-rated health charities accomplish more good per dollar spent. GiveDirectly transfers $1,000 to each household; with $1,000, the Against Malaria Foundation, would purchase more than 150 bednets to protect more than 250 people from malaria. We would not guess that a given household can derive as much benefit-per-dollar from $1,000 as it can from the ~$10-$20 needed to supply it with bednets.

GiveDirectly believes it can transfer 92% of total costs to poor households. In its projections, GiveDirectly forecasts that it will spend 3.3% on identifying the household, 2.9% to transfer the funds, 0.5% to follow up with the household, and 1.5% on other operational costs (paid for by the GiveDirectly board).52

Due to GiveDirectly's model of first spending money to identify and verify the eligibility of households and then making grants over time, and because GiveDirectly is a new organization that has not completed many grants, spending to date may not reflect the long-term ratio of non-grant to grant expenses. Of a total $166,031 in total GiveDirectly costs as of April 2012, 76% had reached Kenyan households. If future transfers to households enrolled as of April 2012 are accounted for, GiveDirectly projects that 91% of total spending will reach these households.53 For details, see GiveDirectly's annotated budget projections.

Note that most of the costs of the RCT are excluded from GiveDirectly's budget. GiveDirectly has estimated that the experiment budget has covered $4,000 worth of costs that GiveDirectly would have paid in the absence of the RCT. This $4,000 is included in the costs-to-date figures above.54

Room for more funds?

As of January 2012, GiveDirectly had $67,000 in unallocated funds.55 It expects to use these unallocated funds plus any additional funding it receives to:56

• Provide an additional $700 to each household that is currently assigned to receive $300 as part of the RCT. GiveDirectly would need about $253,000 to fund these additional grants (see above), plus transfer and follow up costs (estimated at 3-4% of the granted amount, see above).
• Provide $1000 grants to RCT control group households. GiveDirectly would need $507,000 to fund these grants.57 It would also incur some costs to audit a sample of these households to check their eligibility, as well as to transfer the funds and follow up with the recipients.
• Identify additional recipients and provide $1000 grants.

GiveDirectly told us, "The prioritization [of the above grant types] will depend on the flow of incoming funds. If we receive large donations (e.g. $500K), we will organize a new round of recruitment and disburse the money to newly-identified households. If we receive smaller amounts, following completion of the RCT we will top-up treatment households that have received $300 to $1K and then to send control households $1K transfers."58

To date, GiveDirectly has used additional funds to increase the size of existing grants from $300 to $1,000 rather than increase the number of recipients who would receive grants.59 GiveDirectly told us that $1000 is its current target and it does not currently expect to increase the size of grants further.60

GiveDirectly has told us that it believes its model would encounter few barriers to rapid scale-up,61 but that it believes it is appropriate to limit its revenues in the short-term to $5 million. Any donations received after it reached $5 million in revenues would be declined until the first $5 million had been distributed.62

Note that the donation cap was not in place at the time we discussed GiveDirectly's capacity and plans on our blog. The cap was instituted, in part, as a result of this discussion.63

Unresolved issues

Our main reservation about GiveDirectly is its lack of a track record. We believe that it has established strong processes for ensuring that transfers are targeted at and reach very poor households and for monitoring the quality of its program. These processes are largely untested, however, and we would want to see more results before awarding GiveDirectly our highest ratings.

In addition, we have not fully answered the following questions:

• Do cash transfers raise recipients' standards of living above and beyond the value of the transfers? How do the benefits of cash transfers compare to those of the best global health programs?
• Does distribution to some members of a community create jealousy among other members of the community (thereby causing potential for negative impacts within the community)?
• Do grants distort incentives for individuals?
• Do grants distort local markets?

Sources

• GiveDirectly. Analysis plan for randomized controlled trial (PPT).
• GiveDirectly. Annotated budget projections (May 11, 2012) (PDF).
• GiveDirectly. Audit log (XLS).
• GiveDirectly. Capacity note (DOC).
• GiveDirectly. Clarifications on GiveWell's draft review of GiveDirectly (DOC).
• GiveDirectly. Contextualizing transfer size (DOC).
• GiveDirectly. FAQs. http://givedirectly.org/faqs.php (accessed January 12, 2012). Archived by WebCite® at http://www.webcitation.org/64e05vHkD.
• GiveDirectly. Financials. http://givedirectly.org/financials.php (accessed January 12, 2012). Archived by WebCite® at http://www.webcitation.org/64e07SFAl.
• GiveDirectly. GiveWell clarifications (March 28, 2012) (DOC).
• GiveDirectly. IRS form 990 (2010) (PDF).
• GiveDirectly. Notes on verification data (November 17, 2011) (PDF).
• GiveDirectly. Offering memorandum (January 2012). GiveDirectly asked us to keep this document confidential.
• GiveDirectly. Overview of targeting process (PDF).
• GiveDirectly. Survey for randomized controlled trial (XLS).
• GiveDirectly. Team. http://givedirectly.org/team.php (accessed on January 12, 2012). Archived by WebCite® at http://www.webcitation.org/64e1DISuU.
• GiveDirectly. Updated data (March 31, 2012) (XLS).
• GiveDirectly. Verification data (November 17, 2011) (XLS).
• GiveDirectly. Verification template (November 7, 2011) (DOC).
• Mukhopadhyay, Piali. GiveDirectly Staff Member. Email to GiveWell, January 25, 2012.
• Niehaus, Paul, and Jeremy Shapiro. GiveDirectly Founders. Phone conversation with GiveWell, October 6, 2011.
• Niehaus, Paul. GiveDirectly Founder. Conversation with GiveWell, April 1, 2012.
• Niehaus, Paul. GiveDirectly Founder. Email to GiveWell, April 23, 2012.

• 1

  "GD incorporated in 2009 but began accepting anonymous donations from the public in July 2011. GD has had one person working full-time since January of 2011. Jeremy Shapiro was full time during 2011 and Piali Mukyopadhyay is full-time now." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 2

  "GiveDirectly plans to send each recipient household $500 per year for two years." GiveDirectly, "Offering Memorandum (January 2011)," Pg 21.

• 3

  GiveDirectly, "FAQs."
  "In more detail, we originally designed the RCT to generate as much information as possible, given our budget constraints, about (i) how households use lump sums v.s. streams, and (ii) how men and women use money differently. We placed less priority on learning how uses of funds varied with the size of the transfer. In subsequent conversations with people outside our team we have realized that others are interested in the latter question, and so we have scaled up transfers to a subset of the households in the RCT to the full $1,000 in order to generate information on this point." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 4

  "Pilot phase: During various early-stage pilots GD sent $11,921 to 367 households; transfer sizes varied around the mean of $32 per household. Let us know if a more detailed breakdown would be helpful." Piali Mukhopadhyay, email to GiveWell, January 25, 2012.
  Other data from GiveDirectly, "Updated Data (March 31, 2012)."

• 5

  GiveDirectly, "Overview of Targeting Process."

• 6

  "2009 Census data available through the recently launched “Kenyan Open Data Initiative” will be used for future region selection. In addition to the overall incidence of poverty in a region, GiveDirectly also considers the following criteria:

  • Population density, which affects the costs of locating recipients
  • The presence of other NGOs assisting the poor
  • Security, which must be sufficient to allow field officers to operate and markets to function

  After choosing regions GiveDirectly identifies villages within these regions for transfers. Since systematic data on poverty rates are not generally available for small geographic units, GiveDirectly gathers information through discussions with local residents, collecting both subjective evaluations of village poverty as well as numerical estimates of the proportion of homes in each village made without cement, stone or iron. GiveDirectly views village-level targeting as an area for improvement in the future and is currently assessing the utility of other metrics such as distance to nearest paved road or market center." GiveDirectly, "Offering Memorandum (January 2012)," Pgs 16-17.
  GiveDirectly notes that it did not succeed at this at its first try: "To assess whether its targeting process identifies poor households, GiveDirectly has analyzed household survey data from Rarieda District, where it currently operates. Rarieda was chosen using Government poverty estimates from 1997, which showed that 74% of individuals in the region were below the poverty line as of that date. Subsequently released data from 2005, however, show much lower average poverty rates in the region. Working in Rarieda was thus a retrospective mistake, putting more pressure on the household-level targeting process to find poor recipients within the region." GiveDirectly, "Offering Memorandum," Pg 23.

• 7

  "After identifying villages GiveDirectly conducts a proxy means test to identify the poorer households within those villages. Specifically, GiveDirectly defines eligible households as those living in homes made from organic materials, i.e. lacking cement, stone or iron....Field visits are unannounced to minimize the possibility that recipients take actions to game the system (for example, temporarily moving into an abandoned mud and thatch home)." GiveDirectly, "Offering Memorandum (January 2012)," Pg 17.

• 8

  "Field staff members give every identified eligible household a new SIM card (costing approximately $0.50) that will enable them to receive electronic transfers. Households without a cell phone are also offered the opportunity to purchase one by having the retail price of the phone deducted from their future transfer. Purchasing a phone is strictly optional, but it does make it easier for the recipient to collect transfers and for GiveDirectly to stay in touch with him/her." GiveDirectly, "Offering Memorandum (January 2012)," Pg 17.

• 9

  "The transfer is made using M-PESA, a mobile money transfer service operated by Safaricom, a Kenyan mobile telephone operator managed by Vodafone Group Plc. There are several steps linking your donation to a Kenyan household. You make a donation through the GiveDirectly website, or send a check, which is deposited in GiveDirectly's US based bank account. These funds are then sent to GiveDirectly's Kenyan bank account via wire transfer, which is linked with our M-PESA account. From that account, GiveDirectly sends electronic money to impoverished households. The recipient receives a text message informing them of the transfer. They can then exchange their 'mobile money' for cash anywhere in the M-PESA agent network. M-PESA agents are often local shopkeepers but others are petrol stations, supermarkets, courier companies, cyber cafes, retail outlets and even banks. An important feature of this system is that individuals can receive transfers even if they do not own a phone themselves: the M-PESA agent simply loads their SIM card into one of his own phones." GiveDirectly, "FAQs."

• 10

  "We have enrolled 504 treatment households and earmarked $300 for each of these for the RCT. To date we have transferred a total of $43,130 to 412 of these households for an average transfer size of $105, with $108,070 remaining to be transferred between now and September 2012. In light of recent donations received we have decided to increase RCT transfers from $300 to $1,000 for a randomly selected subsample of 142 households. The additional transfers, totaling $100,000, are scheduled to begin in February 2012." Piali Mukhopadhyay, email to GiveWell, January 25, 2012.

• 11

  "After identifying villages GiveDirectly conducts a proxy means test to identify the poorer households within those villages. Specifically, GiveDirectly defines eligible households as those living in homes made from organic materials, i.e. lacking cement, stone or iron....Field visits are unannounced to minimize the possibility that recipients take actions to game the system (for example, temporarily moving into an abandoned mud and thatch home)." GiveDirectly, "Offering Memorandum (January 2012)," Pg 17.

• 12

  "GiveDirectly locates poor Kenyan households, targeting those living in mud and thatch homes." GiveDirectly, "Offering Memorandum (January 2012)," Pg 5.
  
  
  "We condition household eligibility on three distinct criteria: the materials used in their roof, the material used in their walls, and the material used in their floor. This distinction is important as households that meet one or two of these criteria are statistically quite different from those that meet them all." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 13

  GiveDirectly, "Survey for Randomized Controlled Trial."

• 14

  GiveDirectly, "Offering Memorandum (January 2012)," Pgs 23-24.

• 15

  "We plan to continue collecting data on poverty indicators as part of our ongoing enrollment process. We are analyzing baseline data from our RCT to identify subsets of variables that are good statistical predictors of broader measures of consumption poverty." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 16

  GiveDirectly, "Offering Memorandum (January 2012)," Pg 25.
  

• 17

  "The research team drew a random sample from the population of households living in different categories of homes." GiveDirectly, "GiveWell Clarifications (March 28, 2012)."

• 18

  "GiveDirectly conducts two rounds of independent verification before sending transfers. In the first round, 100% of identified households are visited by a second field officer, different from the field officer who first identified the households. This field officer verifies that the enumerated household lives in the specified home, meets the eligibility requirements, and received a SIM card from a GiveDirectly representative, and that the phone number on that SIM card matches GiveDirectly’s records. The field officer also asks whether anyone tried to extract a bribe from the recipient or to coerce them into sharing transfers. In the second round, a GiveDirectly officer (a board member or foreign employee) audits at least 10% of enrolled households, asking the same questions as the field officer. Field officers are told they will be dismissed if they are found to have enrolled ineligible households or demanded bribes...Of 140 such [second-round] audits conducted to date, 1 (1 percent) detected a case of borderline eligibility. In this case the enrolled recipient could not unambiguously establish that he lived at the house on record, which appeared possibly abandoned." GiveDirectly, "Offering Memorandum (January 2012)," Pgs 18 and 24.

• 19

  GiveDirectly, "GiveWell Clarifications (March 28, 2012).

• 20

  "The information... we recorded in these audits was whether a household was ineligible or had paid a bribe." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 21

  "To assess whether the targeting rule is violated due to fraud, GiveDirectly sends higher-level officers (typically a Board member) to audit a fraction of the enrolled households in person. Of 140 such audits conducted to date, 1 (1 percent) detected a case of borderline eligibility. In this case the enrolled recipient could not unambiguously establish that he lived at the house on record, which appeared possibly abandoned." GiveDirectly, "Offering Memorandum (January 2012)," Pg 24.
  "We found one borderline ineligible case and no reports of bribes." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 22

  Paul Niehaus and Jeremy Shapiro, phone conversation with GiveWell, October 6, 2011.

• 23

  GiveDirectly, "Audit Log."

• 24

  GiveDirectly, "GiveWell Clarifications (March 28, 2012)."

• 25

  "The earliest pilots we did were extremely hands on. The objective in the future is to step back a little and audit a subset, maybe 40%, while tapering down from 100%. We want to find the right level of scrutiny." Paul Niehaus and Jeremy Shapiro, phone conversation with GiveWell, October 6, 2011.

• 26

  GiveDirectly, "Offering Memorandum (January 2012)," Pgs 18-19.

• 27

  GiveDirectly sent 7 versions of the survey forms it used, as the form was modified over time. For the most recent version we have seen, see GiveDirectly, "Verification Template (November 7, 2011)." Data is in GiveDirectly, "Verification Data (November 7, 2011)."

• 28

  "This data comprises the results of successful verification surveys conducted by GiveDirectly from June to November 2011. Successful surveys are defined as those where someone in the recipient household could be reached either in person or by phone unsuccessful phone surveys are common (phones are often switched off to preserve battery life). When verification surveys are not successful, GiveDirectly attempts to reach the household again in the future." GiveDirectly, "Notes on Verification Data (November 17, 2011)."

• 29

  GiveDirectly, "Updated Data."

• 30

  The "Notes" column for each of these participants says, "Took some time to register, has started receiving transfers since verification (according to M Pesa)." GiveDirectly, "Verification Data (November 17, 2011)," Sheet "RCT verifications (Cleaned)."
  GiveDirectly told us, "The issue was that some households had not started receiving transfers because they had not yet registered for M-Pesa, and we later confirmed that they had successfully registered with M-Pesa (which can be done online without a phone call). If the issue were that the M-Pesa agent was being uncooperative (say) then we would have to check on that by calling." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 31

  GiveDirectly, "GiveWell Clarifications (March 28, 2012)."

• 32

  The "Notes" column for this participants says, "GE followed up on reported bribe request from M Pesa agent to inform her of actions she could take - on this follow up it seemed she had been confused about the question and had only paid the M Pesa fee, not a bribe." GiveDirectly, "Verification Data (November 17, 2011)," Sheet "RCT verifications (Cleaned)." We assume "GE" is a typo of "GD."

• 33

  GiveDirectly, "Verification Data (November 17, 2011)."

• 34

  "Households in the RCT are receiving $300 each. Households in the pilot project received less." Paul Niehaus and Jeremy Shapiro, phone conversation with GiveWell, October 6, 2011.

• 35

  GiveDirectly, "FAQs."

• 36

  "In brief, the two things that changed were (a) our funding situation, and (b) our sense of what people outside the organization wanted to learn from the RCT.
  In more detail, we originally designed the RCT to generate as much information as possible, given our budget constraints, about (i) how households use lump sums v.s. streams, and (ii) how men and women use money differently. We placed less priority on learning how uses of funds varied with the size of the transfer. In subsequent conversations with people outside our team we have realized that others are interested in the latter question, and so we have scaled up transfers to a subset of the households in the RCT to the full $1,000 in order to generate information on this point...
  We plan to publish detailed results from the RCT on the uses of $300 v.s. $1000 and then have open conversations both within and outside our team about whether these push us towards a different transfer size." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 37

  GiveDirectly, "Contextualizing Transfer Size."

• 38

  GiveDirectly sent us a template for the full survey it will be using and a plan for analyzing the data:

  • GiveDirectly, "Survey for Randomized Controlled Trial."
  • GiveDirectly, "Analysis Plan for Randomized Controlled Trial."

• 39

  "In light of recent donations received we have decided to increase RCT transfers from $300 to $1,000 for a randomly selected subsample of 142 households. The additional transfers, totaling $100,000, are scheduled to begin in February 2012...We currently have approximately $67,000 in unallocated cash. How we use this will depend on how much money we receive during Q1 of 2012. If demand is high enough to justify the fixed cost of another recruitment trip then we expect to conduct a trip and send $1,000 transfers to these new households. If demand is low we expect to wait until the RCT endline survey concludes and then transfer $1,000 to control households from the RCT." Piali Mukhopadhyay, email to GiveWell, January 25, 2012.

• 40

  "In addition to monitoring potential problems in real time, GiveDirectly is conducting a longer-term evaluation to provide more detailed evidence on how recipients use cash transfers. The study is led by Dr. Johannes Haushofer of Harvard University, coordinated by the external research organization, Innovations for Poverty Action, and funded by the National Institutes of Health. It uses the current state- of-the-art approach to evaluation, the randomized controlled trial (RCT). RCTs work like medical trials: the research team identifies a large number of eligible households and then chooses some of them by lottery to receive transfers. Comparing households that received transfers to those that did not yields unbiased estimates of how transfers affect recipients' lives. The study is currently running in the field and on track to finish by early 2013.

  The study will measure the impacts of transfers on household consumption, hunger and nutrition, the entrepreneurial activities of households, health (both physical and emotional) and whether children are in school. It will also examine effects on levels of the stress hormone cortisol, an innovative new way of quantifying well-being. Finally, the study will monitor whether transfers create tensions within recipient communities as well as any other potential spillover effects, such as stimulating local business activity or local inflation." GiveDirectly, "Offering Memorandum (January 2012)," Pg 26.

• 41

  GiveDirectly, "Offering Memorandum (January 2012)," Pg 21.

• 42

  GiveDirectly, "Verification Data (November 7, 2011)."

• 43

  Line item explained in Paul Niehaus, email to GiveWell, April 23, 2012.

• 44

  GiveDirectly sent 7 versions of the survey forms it used, as the form was modified over time. For the most recent version we have seen, see GiveDirectly, "Verification Template (November 7, 2011)." Data is in GiveDirectly, "Verification Data (November 17, 2011)."

• 45

  GiveDirectly, "Verification data (November 17, 2011)."

• 46

  "GiveDirectly stops issuing transfers after two years and clearly informs recipients that they should expect this." GiveDirectly, "Offering Memorandum (January 2012)," Pg 33.

• 47

  GiveDirectly, "Offering Memorandum (January 2012)," Pg 24.

• 48

  Paul Niehaus and Jeremy Shapiro, phone conversation with GiveWell, October 6, 2011.

• 49

  "As of the end of January we have attempted to reach 70% of the households that have received transfers, and have successfully contacted 78% of these. Calls to the remainder of the recipients are in progress. We are weighing conducting an in-person follow-up with a sample of those we ultimately do not reach by phone...
  Our policy is to interview all participants who can be reached by phone or by calling a neighbors phone, and we are weighing conducting an in-person follow-up with a sample of those we have not reached." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 50

  "Did not receive transfer from GD at time of verification: 7.47%.
  There are some delays (takes people time to get ID, sometimes register in someone elses name and issue must be corrected). As of Oct 24, 2011; only 1 of the 13 are still not receiving transfers."
  GiveDirectly, "Verification Data (November 17, 2011)."

• 51

  GiveDirectly, "Offering Memorandum (January 2012)," Pg 30.

• 52

  GiveDirectly, "Annotated Budget Projections (May 11, 2012)."

• 53

  Data from GiveDirectly, "Annotated Budget Projections (May 11, 2012)."

  	Expenses incurred	Future expenses for currently enrolled households	Total Cost	Per HH	% of Total
  	Expenses incurred	Future expenses for currently enrolled households	Total Cost	Per HH	% of Total
  Transfers made	$129,771	$403,229	$533,000	$1000	91.4%
  Identification costs	$19,024	-	$19,024	$36	3.3%
  Transfer costs	$4,351	$11,973	$16,323	$31	2.8%
  Follow-up costs	$1,053	$1,613	$2,665	$5	0.5%
  Set up and Marketing Costs (paid for by GD Board)	$11,833	-	$11,833	$22	2.0%
  Total spend	$166,032	$416,815	$582,845	$1094	100.0%

• 54

  "[We have] included on our books our best estimate of the RCT costs that would normally have been incurred by GiveDirectly, i.e. $4,000, primarily for travel expenses. This cost is included in all of our performance figures." GiveDirectly, "GiveWell Clarifications (March 28, 2012)."

• 55

  Piali Mukhopadhyay, email to GiveWell, January 25, 2012.

• 56

  "Additional funding will be sent to one of the following groups of recipients:

  • Recipients currently in the RCT treatment arm. Because some of these are currently scheduled to get $300, it will take approximately $250,000 to get all of them to $1,000.
  • Recipients currently in the RCT control arm. There are roughly 500 of these and thus capacity for approximately $500,000 in transfers here. Before sending money to this group we would need to re-establish contact with them and audit a fraction of them.
  • Completely new recipients. We would locate these using our standard enrollment process."

  GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 57

  "The total number of control households is 507. Please let me know if I can provide any other information." Piali Mukhopadhyay, email to GiveWell, February 2, 2012.

• 58

  GiveDirectly, "GiveWell Clarifications (March 28, 2012)."

• 59

  "In light of recent donations received we have decided to increase RCT transfers from $300 to $1,000 for a randomly selected subsample of 142 households." Piali Mukhopadhyay, email to GiveWell, January 25, 2012.

• 60

  "We [increased transfer size] because we have made a public statement that we will give each household $1,000, and it is important to us to follow through on that. We had already paid for the smaller $300 grants out of our own pockets." GiveDirectly, "Clarifications on GiveWell's Draft Review of GiveDirectly."

• 61

  "By leveraging the existing M-Pesa infrastructure, GiveDirectly is able to expand capacity to make transfers to poor households simply and quickly. There is no infrastructure to build or partner NGOs to negotiate with - the only requirement is that GiveDirectly identify poor households and provide them with SIM cards.

  To accomplish this, GiveDirectly assembles teams of local field officers for identification. On average, each field worker can enroll 3 households per house, or 75 households per week after allowing for time required for travel between villages. One field manager oversees each team of 5 field officers. Along with back-checkers, this team can typically identify and confirm 1,000 recipients, or enough for $1M of transfers, in 3-4 weeks. Enrolling 4,500 households, implying capacity for $4.5M in transfers, would take an estimates 3-4 months." GiveDirectly, "Offering Memorandum (January 25, 2012)."

• 62

  "Additional donations will be declined until the first $5 million in transfers is successfully completed. This figure reflects the Organization’s assessment of an appropriate rate of scale-up, rather than any organizational need...on average, each field worker can enroll 3 households per hour, or 75 households per week after allowing for time required for travel between villages. One field manager oversees each team of 5 field officers. Along with back-checkers, this team can typically identify and confirm 1,000 recipients, or enough for $1M of transfers, in 3-4 weeks. Enrolling 4,500 households, implying capacity for $4.5M in transfers, would take an estimated 3-4 months." GiveDirectly, "Offering Memorandum (January 2012)," Pgs 6 and 20.

• 63

  "This figure also aligns reasonably well with indications of donor interest. Initial discussions with donors signaled potential interest in this range. GiveWell – a prominent non-profit evaluator – has publicly argued for the importance of distributing a few million dollars to validate the model. According to GiveWell: 'We will become much more confident in GiveDirectly if it executes over the next couple of years, succeeds in meeting its commitment of 90% of expenses given out as cash, and continues to self-monitor and publish its data.'" GiveDirectly, "Offering Memorandum (January 2012)," Pg 30.